CN-121768081-B - AI parallel bar motion teaching system and method

CN121768081BCN 121768081 BCN121768081 BCN 121768081BCN-121768081-B

Abstract

The invention relates to the technical field of intersection of artificial intelligence and physical education, in particular to an AI parallel bar motion teaching system and method. The system comprises a parallel bar positioning sub-module, a protection pad detection sub-module, a judging module and an ROI (region of interest) region detection module, wherein the parallel bar positioning sub-module is used for identifying 2D coordinates of parallel bars by adopting a neural network model, calculating whether the parallel bars incline in a safety threshold range or not, the protection pad detection sub-module is used for detecting a coverage area of a protection pad by utilizing a semantic segmentation network to judge whether the coverage area is in the safety threshold range or not, the judging module is used for judging whether environmental safety detection passes or not, and the ROI region detection module is used for confirming an ROI region in image acquisition of a parallel bar test link. The invention provides an AI parallel bar motion teaching system and method, which comprises the steps of environment safety detection, ROI region detection, human body gesture detection, and output standard parallel bar motion number and nonstandard motion feedback finished by a user by utilizing a time sequence motion gesture analysis and discrimination module.

Inventors

Hong Xianxue
MA XUEER
LU JIANWEI
LIU XIAOCHAO

Assignees

上海百树有方教育装备有限公司

Dates

Publication Date: 20260508
Application Date: 20260303

Claims (8)

1. An AI parallel bar motion teaching system is characterized by comprising a parallel bar positioning sub-module, a protection pad detection sub-module, a judgment module and an ROI (region of interest) area detection module; The parallel bars positioning sub-module is used for identifying 2D coordinates of the parallel bars by adopting a neural network model, obtaining 3D coordinates of the parallel bars by combining information provided by the binocular/depth cameras, and calculating whether the parallel bars incline or not within a safety threshold range; The protection pad detection submodule is used for detecting the coverage area of the protection pad by utilizing a semantic segmentation network, calculating the area ratio of the coverage area to a standard protection area, and judging whether the coverage area is within a safety threshold range or not; The judging module is used for judging whether the environmental safety detection is passed or not according to the detection results of the parallel bar positioning sub-module and the protection pad detection sub-module; the ROI area detection module is used for confirming an ROI area in image acquisition of the parallel bar test link; the human body posture detection module is used for detecting the posture of the student, and the human body posture detection module is used for detecting the human body posture by adopting YOLOv-Pose neural network to generate a skeleton map of the human body.
2. An AI parallel bar motion teaching method for an AI parallel bar motion teaching system according to claim 1, characterized by comprising the following steps: S1, calculating whether parallel bars incline in a safety threshold range or not according to information acquired by a parallel bar positioning sub-module; s2, judging whether the area ratio is within a safety threshold range or not according to the area ratio calculated by the protection pad detection submodule; s3, judging a target student of the parallel bar test according to the acquisition of the image by the ROI area detection module; and S4, conveying the normalized coordinates of different joints into a time sequence action judging module, and judging the score of the parallel bar action.
3. The AI parallel bar motion teaching method according to claim 2, wherein in step S1, the steps are as follows: a. Collecting a parallel bar picture, marking a target frame and two end points of the parallel bars as input, and outputting a detection frame and two end points of the parallel bars after fine adjustment training based on Yolov-pose models; b. internal parameters and external parameters of binocular cameras are known, and stereo correction is carried out on images generated by left and right binocular cameras; c. inputting the corrected left image and right image into a fine-tuning training model, and obtaining four endpoints of parallel bars from each image; d. Performing triangularization operation on each point in the left image and the right image to obtain 3D points; e. after four end point coordinates are obtained, calculating the direction vectors of the parallel bars; f. d, performing three-dimensional reconstruction on the whole scene through the triangulation operation in the step, performing plane fitting on the point cloud of the whole scene, and extracting the normal vector of the ground ; G. calculating parallel bar direction vectors And ground normal vector Is included in the bearing.
4. The AI parallel bar motion teaching method according to claim 3, wherein in step c, four end points of the parallel bars are left end points of a first bar in the binocular left view And right side end point Left end point of second bar And right side end point Left end point of first bar in binocular right graph And right side end point Left end point of second bar And right side end point 。
5. The AI parallel bar motion teaching method according to claim 3, wherein in step d, the left end point of the first bar is taken as an example, and the input pair of points is that And Calculating parallax between two points Then the parallax is converted into three-dimensional coordinates by using the principle of triangulation Repeating the steps to obtain the three-dimensional coordinates of the other three endpoints of the parallel bars.
6. The AI parallel bar motion teaching method according to claim 3, wherein after four end point coordinates are obtained in the step e, a parallel bar direction vector is calculated, and the expression is as follows: 。
7. The AI parallel bar motion teaching method according to claim 2, wherein in step S2, the steps are as follows: A. The method comprises the steps of obtaining 3D coordinates of four endpoints of a parallel bar and a normal vector of the ground, projecting the 3D coordinates of the four endpoints of the parallel bar onto the ground, connecting the four projection points to form a rectangle, and taking the distance of 0.6-1 meter of outward expansion of the rectangle as a standard protection rectangular area A; B. b, inputting a left picture of an image, dividing a protection pad area by using a U-Net network, calculating an external rectangular frame to obtain 3D coordinates of four vertexes of the rectangular frame, and obtaining a projection rectangle B of the rectangular frame on the ground by using the method in the step A; C. and calculating the superposition area of the protection area rectangle A and the protection pad area rectangle B, calculating the ratio of the superposition area to the area of the protection area rectangle A, and judging whether the protection area rectangle A and the protection pad area rectangle B are safe or not.
8. The AI parallel bar motion teaching method according to claim 7, wherein in step S3, the ROI area detection module is configured to confirm the ROI area in the image acquisition of the parallel bar test link, and the steps are as follows: 1) Utilizing Yolov face detection network to find out multiple faces in the image; 2) Presuming the position of the human body according to the position of the human face; 3) And calculating the superposition area of the rectangular projection and the parallel bars, and finding out the bounding box projection with the largest superposition area, namely the target student who will perform the parallel bar test.

Description

AI parallel bar motion teaching system and method Technical Field The invention relates to the technical field of intersection of artificial intelligence and physical education, in particular to an AI parallel bar motion teaching system and method. Background In recent years, physical education is increasingly emphasized, and physical ability testing is also an indispensable test for students. With the development of AI technology, there are some sports motion detection methods based on AI technology, which are applied to horizontal bar, broadcast gymnastics and other sports. However, the prior art has the following defects: The equipment dependence is complex, multiple sensors such as motion capture equipment and an infrared camera are required to be combined, the deployment cost is high, as an intelligent training system and method of Chinese patent publication No. CN109011508A specifically acquire user motion gesture information and vital sign index data in user motion, a motion model can be established, real-time comparison and analysis can be performed, training evaluation and suggestion can be performed on the user motion, and therefore the user motion can be trained and guided in real time on the premise of no need of manual participation; The prior system does not consider the interference of limb shielding (such as the trunk is shielded when the two hands hold the bars) and simultaneous training of multiple people in parallel bar movement, such as an intelligent guidance system applied to pull-up in Chinese patent publication No. CN113694501A, specifically, after capturing real-time motion data of a user in pull-up, the real-time motion data are analyzed according to stored standard motion data, whether the motion of the user is standard or not is judged, and the user is guided to perform motion adjustment through a display enhancement technology when the motion of the user is not standard, so that the user can timely find and correct errors on the motion of the user during training, the self training effect is improved, meanwhile, the physical damage caused by the action is avoided, and the physical health and safety of the user are ensured; the safety detection is missing, namely intelligent detection is not carried out on safety elements such as a parallel bar fixed state, a protection pad arrangement and the like; The algorithm has insufficient robustness, the traditional posture comparison algorithm is easily affected by visual angle difference (such as misjudgment caused by the fact that a standard action library cannot cover individual body type difference), such as a comprehensive monitoring method and system for the movement capability of Chinese patent publication No. CN110464356A and an intelligent training system and method of Chinese patent publication No. CN 109011508A; Therefore, the AI parallel bar motion teaching system and method are designed, and are mainly oriented to intelligent detection of parallel bar motion, wherein parallel bar motion comprises front and back swing, front swing, arm bending and stretching and the like. Traditional physical ability test relies on teacher's manual judgment, and parallel bars action is faster, and manual judgment and count have erroneous judgement and error easily, and efficiency is lower and work load is great. The existing AI parallel bar detection method only detects the bending and stretching of the parallel bars (an automatic test method, device, equipment and medium for bending and stretching of the parallel bars of chinese patent publication No. CN 116392798B), but the lower body motion determination in the parallel bars is more difficult due to the problems of shielding and the like, and the current work lacks in-depth analysis and research. Disclosure of Invention Based on this, it is necessary to provide an AI parallel bar motion teaching system and method for solving the technical problems set forth in the background art. In order to solve the technical problems, the invention adopts the following technical scheme: An AI parallel bar motion teaching system comprises a parallel bar positioning sub-module, a protection pad detection sub-module, a judgment module and an ROI (Region of interest, chinese is an interested area) detection module; The parallel bars positioning sub-module is used for identifying 2D coordinates of the parallel bars by adopting a neural network model, obtaining 3D coordinates of the parallel bars by combining information provided by the binocular/depth cameras, and calculating whether the parallel bars incline or not within a safety threshold range; The protection pad detection submodule is used for detecting the coverage area of the protection pad by utilizing a semantic segmentation network, calculating the area ratio of the coverage area to a standard protection area, and judging whether the coverage area is within a safety threshold range or not; The judging module is used for judging whether the environm